Cartoon Spider Walking a Dog in the Rain and Other AI Generated Images

[Drakkith]

So I just discovered DALL*E, an AI that generates images based on text descriptions input by the user.
And it is FANTASTIC. Fantastically amusing, that is. The images are generated in blocks of 9, and are VERY strange.
The full version is a commercial product I believe, but there's a 'mini' version that generates lower resolution images that can be found here: https://huggingface.co/spaces/dalle-mini/dalle-mini

Here's a screenshot of the images generated by the input "cartoon spider walking a dog in the rain". I haven't checked to see if the same input generates different images each time.

Try the program out and post your favorite here.

 

[Drakkith]

Darth Vader eating sushi in bed:

 

[Borg]

Quantum Mechanics:

Black Hole:

 

[Borg]

"rama lama ding dong" returned a bunch of Dali Lama pictures. No Ding Dongs though.

 

[fluidistic]

I tested it, no luck.
Mario bitcoin running produced the horrible pictures:
Then I tried "Scorpion cleaning a mouse", and there was no scorpion, only mouses. One of which had 3 ears... but nothing like a scorpion anywhere. I quit.

 

[fresh_42]

I only thought this one was necessary for the sake of completion:

Schrödinger's cat

 

[Borg]

Physics Forums:

 

[Borg]

I tested it, no luck.
Mario bitcoin running produced the horrible pictures:
Then I tried "Scorpion cleaning a mouse", and there was no scorpion, only mouses. One of which had 3 ears... but nothing like a scorpion anywhere. I quit.

Don't give up. I've been running dozens and have found the following:

• Pictures of people are terrible but inanimate things like trees are quite interesting.
• Purposeful mispellings tend to come back interesting. I tried "all schtook up" and it had nothing to do with Elvis.
• It's Dali, so the weirder the phrase, the better. I tried treebeard and got a LOTR set of pictures. Then I tried treebard which put mouths on trees. Finally, the phrase "tree barge" returned barges with trees growing out of them. Some of the pictures in "Flying trees" looked like they had rocket engines pushing them skyward.

Another set using the phrase "dali eating the moon". Very Dali-like:

 

[strangerep]

"rama lama ding dong" returned a bunch of Dali Lama pictures. No Ding Dongs though.

I dunno. Can you prove that the Dali Lama is not a Ding Dong? I find it more likely than not.

 

[robphy]

 

[Drakkith]

 

[berkeman]

I only thought this one was necessary for the sake of completion:

Schrödinger's cat

But, but, they're all alive! That can't possibly be right...

 

[Oldman too]

But, but, they're all alive! That can't possibly be right...

I copy/pasted that line into the bar, the resulting images were too bizarre to post.
This is an interesting paper, thought I'd drop the link before posting my efforts at AI generated images. https://arxiv.org/pdf/2206.00169.pdf

Relativity.

GPT's face.

MC Escher (not too bad a job on the tessellations)

 

[DennisN]

So I just discovered DALL*E, an AI that generates images based on text descriptions input by the user.
And it is FANTASTIC. Fantastically amusing, that is.

Yes, it's very fun!
Me and my friends have sent each other funny images from DALL-E mini for some time.
Here are some of my creations (quite bizarre ):

Girl with a Pearl Earring smoking:

Mona Lisa using a Macintosh computer:

Jesus demonstrating a Macintosh computer:

Furthermore, one of my friends recently sent me a number of fantastic high resolution images made with DALL-E (I don't think it's possible at this moment to generate hi-res images with DALL-E online):

A cat with a pearl earring:

Cat on chessboard:

Cathulhu:

Cleopatra checking her iPhone:

Dog wearing a hot dog suit:

Garfield horror film:

Girl with a pearl earring behind the scenes:

(to be continued...)

 

[DennisN]

(...continued)

Homer Simpson running in a mall:

Mc End:

Portable washing machine for cats:

 

[Drakkith]

@DennisN Wow, those are amazing pictures. Wish I could use the good version!

 

[DennisN]

Wish I could use the good version!

Yeah, me too!

 

[Drakkith]

For comparison, here's the 'dog in hot dog suit' from the Dall E mini version:

 

[DennisN]

I generated a brand new one today:

Darth Vader drinking Coca Cola
("You don't know the power of the dark coke...")

 

[Oldman too]

Yeah, me too!

Some of your images have exceptional resolution, do some of your cohorts have the commercial version? The search parameters your using are great by the way.
(edit note) I see some like the McEnd are done on DALLE2, got to look into that.

This is an idea my wife came up with for one.

 

[Oldman too]

It looks like a put your name on the list, situation.
https://openai.com/dall-e-2/

https://arxiv.org/abs/2204.06125

https://arxiv.org/pdf/2204.06125.pdf

 

[DennisN]

Darth Vader drinking Coca Cola

There must be balance in the Force, so here's "Yoda drinking a Pepsi":

 

[Drakkith]

Just got through the waitlist and was able to use the full program just now. It's much better than the free version, but certainly not perfect. For comparison, here's the 'cartoon spider walking a dog in the rain' prompt in the full version. Compare to the original up in the first post.

Here's 'darth vader eating sushi in bed'. Compare to original up in the 2nd post.

Here's 'digital art of darth vader fighting batman on a train in the moonlight'.

Note that the original images are full size. These are just a quick screenshots.

 

[Drakkith]

Oh wow! Easily my favorite so far!
Input phrase: 'digital art cute fox scientist mixing chemicals'

 

[Drakkith]

Phrase: watercolor mother holding baby in the sunset

@Evo here's a cute kitty or four for you.
Phrase: painting of cute kitten in grass

Phrase: digital art of a dragon painting a picture of a puppy

That's probably all for now unless I see some that really stand out. I've got about 40 credits left and each phrase has cost me 1 credit so far to generate 4 images, so I can take a couple of requests. Credits currently cost $15 for blocks of 115 credits, but I got 50 for free initially.

 

[Drakkith]

Okay, the previous post was supposed to be the last for tonight, but I just found out it can do writing/text! Wanted to send my fiancée something cute so I put in this phrase: digital art cute red panda with phrase 'I Love You' above head

It even put the heart in place of the word 'love' in the 2nd picture!

 

[Hornbein]

Don't give up. I've been running dozens and have found the following:

• Pictures of people are terrible but inanimate things like trees are quite interesting.

It's the free version. People are uglified deliberately.

 

[Drakkith]

It's the free version. People are uglified deliberately.

Are you certain of that?

 

[Hornbein]

Are you certain of that?

Yeah. I frequent an artists' group. Some are willing to pay for the real thing and show the results.

If people can get the service for free then most will never pay.

 

[Drakkith]

Yeah. I frequent an artists' group. Some are willing to pay for the real thing and show the results.

If people can get the service for free then most will never pay.

I assumed that the free version was just much worse overall at everything, including making faces, than the paid version. For the reasons you stated.

 

[jack action]

Rabbits against magic has made today's comics strip with this program (or a similar one):

 

[Swamp Thing]

"Very cute, but not for the faint-hearted"

 

[Swamp Thing]

 

[Swamp Thing]

 

[Office_Shredder]

This seems like a relevant update to the thread

https://www.cnn.com/2022/09/03/tech/ai-art-fair-winner-controversy/index.html